1:1 ABM pages is the easy part

I work on AI Innovation here at Optimizely. For the last 6.5 years, we've always been at the forefront of using our own products as "customer zero" — that's what we call ourselves — implementing Experimentation, CMS, and AI into our daily lives. I like to say I have the best job as a marketer in the world, because I get to use all the fun stuff without having to do any procurement. Occasionally it's frustrating. Often, it's hard. But it's always been rewarding. Many of the challenges we face internally are the same ones our customers grapple with. I just don't pay for anything.

I'd love to share some of those learnings with you.

The most surprising thing about last year isn't what we started doing — though we did a lot more. It's what we stopped.

When we were challenged internally with "AI adoption" following a productivity mandate, we scratched our heads. Do we get everyone chatting more with AI? Build automated workflows? When we asked our peers what they actually needed, they said things like: "We'd love to create super hyper-customized swag at every booth."

Really? That's the biggest bottleneck?

It wasn't.

So we ran a different exercise. We asked each team to walk us through an entire campaign — an event, a social post, whatever. Every step, from start to finish. Then mark each part that was:

• Most frustrating
• Most time-consuming
• Hardest to achieve quality on (I'll come back to that one)

Turns out it wasn't swag. It was things like: "It's hard to get information from that other team," or "We have to wait for an asset to get approved." Anything involving other teams, documents, and processes — that's where the time and frustration lived.

As an aside: try this exercise yourself. I personally find it a bit absurd that people think AI's main job is to replace them. Companies might feel that way. Leadership might feel that way. But I feel busier now than I've ever been! If you use it well, AI should be a multiplier. And the only way to get there is to compress the other work first.

Will there be companies that switch entirely to AI “slop”? Absolutely. But I've found it exceptionally good at making the boring parts of my job obsolete — and I'd much rather focus on that than ask someone else to do it. Nobody can honestly claim their entire day is all fun and games. Unless you work at a pet store. Maybe.

None of that friction was hard work. It was just work someone had to do, and the cost of doing it was that nobody had time for the thinking part. We had normalized it so completely we'd forgotten it was friction. It was just the job.

What I do now is closer to problem definition than execution. Which sounds insufferably abstract right up until you try it for a few months and realize it's the whole game. Most of what used to be "marketing work" was downstream of badly framed problems that nobody had time to reframe, because everyone was too busy doing the badly framed work. Take the doing off the plate, and the framing becomes the work. The framing is harder than the doing ever was.

I came up as an operator. I measured myself by output — campaigns shipped, tests run, pages launched. The new job measures differently. You aren't producing the artifact anymore; you're deciding which artifacts should exist, who they're for, and how each one compounds into the next. Some weeks I miss the certainty of just shipping something.

Here's a concrete example.

We ran a competitive campaign across about 2,000 accounts. Each one got an individualized landing page — different headline, different proof points, different competitive framing — built from contextual signals about that specific company. Recent job postings. Press releases. Leadership changes. Product announcements. The kind of personalization that, a year ago, we reserved for maybe twenty named accounts, because the manual research and content production cost too much to scale further.

This wasn't 10x faster. It was impossible before. Not impossible in the rhetorical sense, but impossible in the literal sense that no headcount or budget I could realistically defend would have paid for the work. We weren't choosing between fast and slow. We were choosing whether entire portions of our market got individualized treatment at all.

But here's what nobody talks about enough: the number wasn't the point. Everybody and their uncle can now claim they made 2,000 pages. What mattered was that those pages were on-brand, performant, and compliant. That when we needed to change something, we could update them at scale without a week-long cleanup sprint. That we had systems to build it, manage it, and report on it, and none of those systems existed before.

We weren't trying to duplicate our website overnight. We were trying to build rich, high-quality experiences that we actually owned and could govern. That's a different problem. And it's the one most teams skip straight past when they get excited about volume.

The interesting part isn't that we did it. It's what happened after. Once you can personalize at that scale, the bottleneck moves. The question stops being "can we afford to personalize?" and becomes "is the signal we're personalizing on actually any good?" Which is a much harder, much more strategic question. The constraint shifts up the stack. You go from solving content production to solving content judgment.

That move — from production to judgment — is the actual change. Everything else is just the mechanism.

I hear this constantly from digital leaders: "I don't want a tool that doubles the size of my website."

And honestly? That's a valid instinct. You should be skeptical of that.

But that's not what's on offer here — at least not if you approach it right. The goal was never more pages for the sake of more pages. I want less content. I want content that's better. The question to ask isn't "how many pages can I generate?" It's "can I build the best version of this, fast, and then actually maintain it?"

Think of it like the agile skateboard-to-car progression except AI gives you six cars now, and you can rocket-ship straight to the end goal as your starting point instead of iterating incrementally toward it. You still iterate. You just iterate off of something far better than what you had before.

The concern about losing control is real, though. Does this mean I'm handing the keys to my website to everyone? In an enterprise context, that's not nothing. There are brand guidelines, legal approvals, accessibility requirements, and performance standards to uphold.

Here's how I think about it. A well-designed agent should follow a simple progression: 

• identify
• recommend
• resolve.

First, it identifies the problem. All your product images are missing alt tags. Your page descriptions are out of date. Here are the pages that are out of compliance with brand guidelines. This is pure signal. No action, just clarity.

Next, it recommends. Here's the updated alt text. Here's how I'd rewrite those descriptions. Here's a draft that brings those pages in line with the new guidelines. You see it before anything happens. You tweak, you approve, you build trust.

Then — and only once you've built that trust — it resolves. It fixes the broken link. It stages the updated pages. It applies the recommendation. And even then, "resolve" doesn't have to mean "publish." It might mean "put it in staging for my review." You stay in control of how far down the chain the agent goes.

The more AI-mature organizations I see, the more they lean into this framework, and the more autonomy they comfortably extend to agents over time. Not because they stopped caring about quality, but because they built systems that made quality the default.

Plenty. Here's the one that mattered most.

I assumed the right move was to democratize agent-building across the team. Seemed obvious: AI is a force multiplier, everyone should have it, let a thousand flowers bloom.

What we got was pilot sprawl. Disconnected agents that demonstrated novelty without durability. Agents that worked for one person and couldn't be operated by anyone else. Governance gaps that would have become real risks if we'd kept going. We hadn't removed the old constraints. We'd built new ones, with a more impressive vocabulary.

Here's the reframe that helped me: a good email agent isn't for the person who owns email. It's for everyone else. If someone on the content team needs to spin up a Marketo email and doesn't know how — and frankly, shouldn't have to know how — the agent handles 80% of it and stages the rest for the person who does. That's not democratizing tool access. That's democratizing outcomes.

What I had to unlearn was the assumption that AI removes the need for product discipline. It doesn't. It raises the stakes on it. When anyone can build an agent in an afternoon, the question isn't whether to build, it’s which problems are worth solving, and how the solutions stay consistent when the person who built them moves on.

We pulled agent creation into a governed backlog. We ran it through a workshop that diagnoses the real operational constraint before anyone builds anything. We applied the same product rigor to our internal agents that we'd apply to something we'd ship to a customer.

The thousand-flowers approach felt empowering. It produced a thousand stalled experiments.

The shift was treating problem identification as the actual work, and agent-building as the trivially easy part downstream of it. That inversion is harder to make than it sounds, because it means accepting that the most valuable people on your team are the ones doing the least visible work — and that the visible output you used to celebrate is now the cheap part.

If you're worried about whether your team is ready for this, I want to be direct: the question is real, but it's not the question you think it is.

The skill gap isn't the problem. Most marketing teams have the technical aptitude to work with AI tools — the tools are getting easier, not harder, and they will keep getting easier. The harder shift is psychological.

The people on my team who struggled most weren't the ones who didn't understand the technology. They were the ones whose identity was tied to the work that got automated. If you measure yourself by how many decks you ship and the decks start writing themselves, you have to find a new measure. If your value was being the person who could pull the report fast and the report pulls itself, you have to redefine what your value is. That's not a training problem. It's not a tooling problem. It's an identity problem, and you can't solve it with a workshop.

So here's what I'd tell a peer leader: your team is not going to be ready in the way you mean. They're going to have to redefine what their work is, in flight, while you redefine what yours is at the same time. Some people will love that. Some will resent it. A few will leave. That isn't a failure of change management. That's the shape of what's actually happening.

The way through it isn't to wait until the team is ready. It's to be honest, early, about what's changing. Don't sell it as "this will make your life easier." It will, eventually. First, it makes your job different, and not everyone signs up for a different job. Treat that conversation as the real work, because it is.

Here's the frame I keep coming back to: nobody can convince me they love every single part of their job every day. There are parts you love and parts you tolerate. Give the parts you tolerate to AI. Focus relentlessly on the parts where you add value that no agent can replicate. The content strategist who loves interviewing people and building narratives? She creates richer, more editorial work than any agent ever will. The video person who lights up when he's ideating and filming? He shouldn't be the one editing and distributing. That's the agent's job. Build systems for the parts people don't love and protect the time for the parts they do.

The agents, the platform, the workflows — those are the easy part. Genuinely. The hard part is the conversation you have with your team about what good marketing work even means now. Have that conversation honestly, and your team will figure out the rest. Avoid it, and no platform in the world will save you.

That's the part I wish someone had told me a year ago.